Probabilistic Flood Forecasting for Small Catchments using the G2G - - PowerPoint PPT Presentation

Probabilistic Flood Forecasting for Small Catchments using the G2G Model

Steve Cole, Alice Robson, Phil Howard, Vicky Bell and Bob Moore

Centre for Ecology & Hydrology, Wallingford

Distributed Model (G2G) Lumped Model

Lumped and Distributed hydrological modelling

Gauging

Hydrological modelling

• One model for each gauging station
• Many parameters calibrated to
• bserved flow location
• Flow estimates for one location only
• Uses catchment average rainfall

Gauging station

• One model for large regions (UK)
• Small set of regional parameters,

strong support from digital datasets

• Flow estimates in each grid (1km2)
• Uses gridded rainfall estimates

Need for distributed models of flood response

Storm total

1

Flow m3s-1 Lumped Model Distributed (G2G) Model

― Catchment-wide storm

Impact of spatial extent and location

• f storm on flood response?

mm hr-1 mm Flow m3s-1 Time (days) Hyetograph

― Lower catchment storm ― Upper catchment storm

Moore et al. (2006), IAHS

• Pub. 305

Darwen at Blue

• Br. (135km2)

• Distributed hydrological models offer a natural

approach to area-wide flood forecasting that includes small catchments BUT:

• What rainfall estimates and forecasts should be

Science Questions

Motivation

• What rainfall estimates and forecasts should be

used?

• How to formulate area-wide distributed models

for operational use in flood forecasting?

• How do these area-wide models perform at

small gauged and ungauged locations?

Gridded rainfall estimators: examples

• Using Hameldon

Hill radar in North- West England

• Two relatively

River Kent Case Study

River Kent at Sedgwick (212km2)

• Two relatively

steep upland catchments

• Strong topographic

control on flow response

Cole and Moore (2009), AWR Darwen at Blue

• Br. (135km2)

Gridded rainfall estimators: examples

• River Kent catchment, orographic event, 3 Feb 2004

2km Nimrod Raingauge-only Gauge-adjusted Raingauge

River Kent Case Study: Gridded rainfall estimators

15 min totals 15 min totals

• River Darwen catchment, convective event, 14 June 2002

1km raw radar Raingauge-only Gauge-adjusted River gauge

Grid-to-Grid distributed model (G2G)

Surface flow-routing Precipitation Evaporation

River Kent Case Study: G2G model

• Uses digital spatial datasets (e.g. terrain)
• Responds to spatial variation of rainfall input
• Grid-to-Grid routing using Kinematic Wave scheme

Saturation-excess surface runoff Drainage River Subsurface flow-routing Return flow River flow Runoff- producing soil column

G2G routing: use of terrain data

• 1. Flow directions:

apply automated method to 50m DTM to infer 1km flow-paths

• 2. Catchment boundary

delineation: inferred

Elevation m

945

River Kent Case Study: G2G model

!

Sedgwick

!

Sedgwick

!

Sedgwick

delineation: inferred from flow-path directions

• 3. Land/river designation:

drainage area + river length threshold

!

Sedgwick

Victoria Bridge Mint Bridge Sprint Mill Bowston

• 4. Select forecast

locations: gauged or ungauged

Raingauge-only Gauge-adjusted radar

Gridded rainfall estimators

2km radar data

Improved rainfall estimates Validated by hydrological modelling Novel multiquadric surface fitting method (Cole & Moore, 2008)

G2G model assessment of rainfall estimators

Distributed model (G2G) Observed Rating curve maximum

15 min totals

(70 km2) (35 km2) (66 km2) (185 km2)

• 1

Flow m3s-1 River Kent Case Study

Elevation m

945

Victoria Bridge Mint Bridge Sprint Mill Bowston

G2G model assessment at ‘ungauged’ sites

(212 km2)

Flow m3s- Flow m3s-1 Time (days) Observed flow Model flow Model baseflow Upper limit of rating equation

!

Sedgwick

!

Sedgwick

!

Sedgwick

!

Sedgwick

Victoria Bridge

• G2G model calibrated

at Sedgwick only

• 15-min raingauge data

used

• Comparable results at

ungauged sites

Hydrological flood forecasts using NWP

Collaboration with the Joint Centre for Mesoscale Meteorology, EA and CEH using the Carlisle 2005 Floods (uses the PDM model). Q: Can new 1 or 4 km NWP rainfalls provide reliable flood forecasts? A: Yes, for the Carlisle floods (orographically enhanced frontal rain)

Flood forecasting using high-resolution NWP 1 km 4 km 12 km (2005) 4 km (2008) 1 km (2011) 12 km

Caldew (246km2)

Time (hours) Flow m3s-1

Raingauge

Flood Warning Level

Roberts, Cole et al., 2009, Met. Apps

12 km 4 km NIMROD radar

Flood forecasting using high-resolution NWP

Boscastle 2004 case study

20 km radius from Boscastle Forecasts from 03 UTC

Courtesy Nigel Roberts, JCMM (Met Office)

12 km 1 km NIMROD radar

Flood forecasting using high-resolution NWP

Boscastle 2004 case study

• 1 or 4km NWP major improvement over 12km product
• Still uncertainty in NWP rainfall intensities and location

20 km radius from Boscastle Peak accumulations up to 50mm

Courtesy Nigel Roberts, JCMM (Met Office)

Forecasts from 03 UTC

Ensemble Flood Forecasting using G2G

1km NWP pseudo-ensemble G2G Model 1km river flow ensemble Comparison with river flow observations

• Simple psuedo-ensemble method developed to capture NWP
• uncertainties. Genuine ensembles will be available in 2012(?)

Flood forecasting using high-resolution NWP

Acknowledgements: Collaboration with JCMM (Met Office)

Probability of exceeding a given flow threshold, for a given forecast horizon

Risk Map of flood exceedance using G2G ensembles and Q(T) flow return period grids

This example employs:

Flood forecasting using high-resolution NWP

Acknowledgements: Collaboration with JCMM (Met Office)

920 km2

This example employs:

• NWP 1km rainfall pseudo-

ensemble

• 10 year return period flow

thresholds

• 24 hour forecast horizon

Potential to identify flood risk hotspots

STEPS 6-hour spatial rainfall forecast 0900 to 1500 20 July 2007 20 ensembles Avon & Tame (Midlands) catchments

Ensemble 1 Ensemble 2 Ensemble 3 Ensemble Average

Midlands Case Study – 20 July 2007

HyradK raingauge STEPS 6 hrs Zero rainfall (padding out)

130 km2

Radar Composite HyradK raingauge

Ensemble average rainfall is less than raingauge rainfall but higher than radar Ensemble hydrographs

• bserved flow

modelled flow using raingauge rainfall 20 flow ensembles using STEPS zero rainfall used beyond 6hr STEPS

T+0 T+6

130 km2

20 6-hour STEPS ensemble rainfall forecasts in G2G 20 July 2007 Observed Modelled 20 STEPS ensemble members

G2G ensembles using STEPS forecasts

Midlands Case Study – 20 July 2007

91 km2 93 km2 185 km2 93 km2

members 20 July 2007 Traditional ensemble

• utputs at

gauged locations

130 km2 74 km2

75-100%

10 year return period flood threshold 6-hr STEPS forecasts then zero rainfall 20 STEPS Members 09:00 20 July 2007 origin Avon & Tame (Midlands) catchments

T+3 hours T+6 hours

Probability of exceedance flood maps

Midlands Case Study – 20 July 2007

Key indicates probabilities of (number of members) exceeding the 10-year flood. During early part of storm, highest exceedance probabilities are on the very small rivers. As time progresses the main exceedance hotspots are on the larger rivers and can be tracked moving downstream and meeting at confluences.

75-100% 50-75% 25-50% 10-25% 2-10%

T+12 hours T+18 hours

• Several EA/Defra R&D projects recommended nationwide
• perational trial of G2G for flood forecasting

– 2004-06: Extreme Event Recognition Phase 2 (FD2208) – 2005-07: Rainfall-runoff and other modelling for ungauged/low- benefit locations (SC030227) – 2007-10: Hydrological modelling using convective scale rainfall

National application of G2G

National application of G2G

– 2007-10: Hydrological modelling using convective scale rainfall modelling (SC060087)

• Pitt Review of the Summer 2007 floods

– Environment Agency/Met Office Flood Forecasting Centre (FFC) for England & Wales, opened April 2009 – Scottish Flood Forecasting Service (SFFS) between SEPA/Met Office opened 2010 – G2G now undergoing operational trials in FFC and SFFS

Runoff production key element – needs to reflect heterogeneous soil properties Use of Soil Survey data (HOST, Seismic,

• ther…) to obtain 1km grids of:

G2G runoff production: use of soil property associations

National application of G2G

• water content at field capacity
• residual soil water content
• porosity
• saturation hydraulic conductivity
• horizon depth

Issues: Scale Effective values Lateral properties

Association table links 29 HOST soil classes to soil properties

Bell et al. (2009), JoH Moore et al. (2006), IAHS Pub. 305

G2G national application

National application of G2G

• G2G runs nationally within NFFS/FEWS using a 15 min time-

step and models river flow and soil moisture on a 1km grid

• Ongoing operational trial and assessment

Raingauge-adjusted radar River flow Soil moisture

Examples of catchments with generally good G2G performance January & February 2008

Observed Modelled

93 km2 53 km2 69 km2 93 km2 357 km2

Demonstrates modelling of different flow regimes and catchment sizes with the G2G Model

357 km 352 km2 191 km2 9,962 km2

G2G Operational Use: Data Assimilation

National application of G2G

• State Correction

– slow adjustment that improves long-term baseflow modelling – good quality flow data needed – adjusts upstream soil storages

Simulation + State Correction

• Flow Insertion

– observed flow fed in at each gauged location with good data – permits ARMA forecast correction

• Flow Insertion and Local Parameter Calibration

– Flow insertion allows nested catchments to be calibrated independently of upstream modelling (e.g. use lake outflows) – River routing speed can be calibrated for each sub-catchment

+ State Correction + Flow Insertion

National application of G2G

G2G Model performance by area

50 100 150 200 250 300 350 <100 100 - 250 250-1000 >1000 Frequency Catchment Area (km2)

Gauged catchments in NFFS

• About 45% of the gauged

catchments in NFFS have an area <100km2

• Results for spring 2008 with state

0.2 0.4 0.6 0.8 1 <100 100 - 250 250-1000 >1000 R2 model efficiency (pooled) Catchment Area (km2)

G2G Model Performance

0.2 0.4 0.6 0.8 1 <100 100 - 250 250-1000 >1000 R2 model efficiency (pooled) Catchment Area (km2)

G2G Model Performance

0.2 0.4 0.6 0.8 1 <100 100 - 250 250-1000 >1000 R2 model efficiency (pooled) Catchment Area (km2)

G2G Model Performance

Catchment Area (km2)

• Results for spring 2008 with state

correction on

• No insertion + global params

Larger catchments tend to perform slightly better

• Flow insertion + global params

Most benefit for large catchments

• Flow insertion + local params

Most benefit for small catchments

National application of G2G

G2G Model performance by area

• National G2G performance maps for spring 2008

Area <100km2 All stations R2

Summary and conclusions

• G2G Model:

– sensitive to spatio-temporal structure of storms – shapes flood hydrograph from storm and landscape properties – Q(T) grids allow mapping between G2G flows and flood return periods – indication of severity

• National application of G2G for flood forecasting:

– results show utility for small catchments and performance improves with catchment size – high-resolution (4 or 1km) NWP provides better rainfall and flood forecasts and indicative flood warnings for the next few days – data assimilation greatly improves forecast performance – can produce real-time flood risk maps, if used with ensemble rainfall forecasts: important for small catchments